Models / Qwen3.5 9B
Can I run Qwen3.5 9B locally?
Qwen · 9.7B parameters · dense · up to 256K context · license apache-2.0 · Hugging Face
At Q4_K_M with a 32K context, Qwen3.5 9B needs about 8.2 GB of GPU or unified memory by our estimate. The official checkpoint is 19.3 GB on disk.
Hybrid model: only the full-attention layers keep a growing KV cache; linear/conv layers keep a small fixed state (not counted).
Memory by quantization and context
| Quantization | Weights | Total at 8K | Total at 32K | Total at 128K |
|---|---|---|---|---|
| Q3_K_M (~4.0 bit) | 4.8 GB | 6.2 GB | 7.0 GB | 10.3 GB |
| Q4_K_M (~4.9 bit, popular default) | 5.9 GB | 7.4 GB | 8.2 GB | 11.4 GB |
| Q5_K_M (~5.7 bit) | 6.9 GB | 8.4 GB | 9.2 GB | 12.4 GB |
| Q6_K (~6.6 bit) | 7.9 GB | 9.4 GB | 10.2 GB | 13.5 GB |
| Q8_0 (~8.5 bit, near-lossless) | 10.3 GB | 11.8 GB | 12.6 GB | 15.9 GB |
| F16 / BF16 (16 bit) | 19.3 GB | 21.2 GB | 22.0 GB | 25.2 GB |
Bits per weight from llama.cpp's quantize docs. F16 KV cache. Totals include our runtime-overhead estimate.
Cheapest hardware that fits (Q4_K_M, 32K)
- NVIDIA Tesla V100 16GB (PCIe or SXM2): about €275 (used, ebay.co.uk, 15.5 GB usable)
- Intel Arc A770 16GB: about €302 (new, alternate.de, 15.5 GB usable)
- Intel Arc B580 12GB: about €320 (new, alternate.de, 11.5 GB usable)
- NVIDIA Tesla T4 16GB: about €521 (used, ebay.co.uk, 15.5 GB usable)
- NVIDIA Tesla V100 32GB (PCIe or SXM2): about €621 (used, ebay.co.uk, 31.5 GB usable)
Runs in
| Tool | Platforms | How | Source |
|---|---|---|---|
| Ollama Desktop app + CLI | macOS, Windows, Linux | ollama run qwen3.5:9b | source |
| LM Studio Desktop app | macOS, Windows (x64, arm64), Linux (x64, arm64) | lmstudio-community build: lmstudio-community/Qwen3.5-9B-MLX-4bit | source |
| llama.cpp Inference engine (C/C++) | Linux, Windows, macOS, Android (build) | GGUF: bartowski/Qwen_Qwen3.5-9B-GGUF | source |
| MLX (mlx-lm) Apple framework + CLI | macOS (Apple Silicon only) | MLX conversion: mlx-community/Qwen3.5-9B-MLX-4bit | source |
| KTransformers CPU+GPU hybrid engine | Linux, Windows | named in the official model card | source |
| vLLM Serving engine | Linux | named in the official model card | source |
| SGLang Serving engine | Linux | named in the official model card | source |
Only tools with a public source are listed: an official build (ggml-org, lmstudio-community, mlx-community, bartowski or Unsloth on Hugging Face), the model card, ollama.com or the tool's own README. Apps built on llama.cpp, such as Jan and PocketPal AI, can usually load the same GGUF files. All local apps.
Try other settings
Estimated memory needed8.2 GB
Hybrid model: only the full-attention layers keep a growing KV cache; linear/conv layers keep a small fixed state (not counted).
45 hardware options fit
Cheapest real offer per device from AI Hardware Deals, converted to EUR at approximate rates. Multi-GPU rows multiply the card price.
| Setup | Headroom | ||||
|---|---|---|---|---|---|
| NVIDIA Tesla V100 16GB (PCIe or SXM2) server GPU (passive cooling) | 1 device | 15.5 GB | 7.3 GB | €275 used, ebay.co.uk | n/a |
| Intel Arc A770 16GB consumer GPU | 1 device | 15.5 GB | 7.3 GB | €302 new, alternate.de | n/a |
| Intel Arc B580 12GB consumer GPU | 1 device | 11.5 GB | 3.3 GB | €320 new, alternate.de | n/a |
| NVIDIA Tesla T4 16GB server GPU (passive cooling) | 1 device | 15.5 GB | 7.3 GB | €521 used, ebay.co.uk | n/a |
| NVIDIA Tesla V100 32GB (PCIe or SXM2) server GPU (passive cooling) | 1 device | 31.5 GB | 23.3 GB | €621 used, ebay.co.uk | n/a |
| AMD Radeon RX 7900 XT 20GB consumer GPU | 1 device | 19.5 GB | 11.3 GB | €889 new, mindfactory.de | n/a |
| Intel Arc Pro B60 24GB workstation GPU | 1 device | 23.5 GB | 15.3 GB | €899 new, mindfactory.de | n/a |
| AMD Radeon RX 9070 XT 16GB consumer GPU | 1 device | 15.5 GB | 7.3 GB | €929 new, mindfactory.de | n/a |
| AMD Instinct MI100 32GB server GPU (passive cooling) | 1 device | 31.5 GB | 23.3 GB | €1,130 used, ebay.co.uk | n/a |
| NVIDIA GeForce RTX 3090 Ti 24GB consumer GPU | 1 device | 23.5 GB | 15.3 GB | €1,499 used, kleinanzeigen.de | n/a |
| NVIDIA GeForce RTX 5080 16GB consumer GPU | 1 device | 15.5 GB | 7.3 GB | €1,664 new, alternate.de | n/a |
| NVIDIA L4 24GB server GPU (passive cooling) | 1 device | 23.5 GB | 15.3 GB | €1,841 used, ebay.fr | n/a |
| NVIDIA A10 24GB server GPU (passive cooling) | 1 device | 23.5 GB | 15.3 GB | €2,253 new, ebay.co.uk | n/a |
| NVIDIA RTX A5000 24GB workstation GPU | 1 device | 23.5 GB | 15.3 GB | €2,348 used, ebay.co.uk | n/a |
| AMD Radeon Pro W7800 32GB workstation GPU | 1 device | 31.5 GB | 23.3 GB | €2,415 new, grafikkarten.com | n/a |
| NVIDIA GeForce RTX 3090 24GB consumer GPU | 1 device | 23.5 GB | 15.3 GB | €2,599 used, alternate.de | n/a |
| Framework Desktop (Ryzen AI Max+ 395, 128GB) unified memory | 1 device | 120 GB | 112 GB | €2,875 used, ebay.co.uk | ≤ 43 tok/s |
| NVIDIA GeForce RTX 4090 24GB consumer GPU | 1 device | 23.5 GB | 15.3 GB | €2,900 new, idealo.de | n/a |
| GMKtec EVO-X2 (Ryzen AI Max+ 395, 128GB) unified memory | 1 device | 120 GB | 112 GB | €3,175 new, newegg.com | ≤ 43 tok/s |
| Apple Mac mini M5 Pro 64GB (2026) unified memory | 1 device | 48.0 GB | 39.8 GB | €3,224 new, cyberport.de | ≤ 52 tok/s |
| NVIDIA A100 40GB PCIe server GPU (passive cooling) | 1 device | 39.5 GB | 31.3 GB | €3,548 new, ebay.com.au | n/a |
| ASUS Ascent GX10 (GB10, 128GB) unified memory | 1 device | 120 GB | 112 GB | €4,136 new, alternate.de | ≤ 46 tok/s |
| NVIDIA A40 48GB server GPU (passive cooling) | 1 device | 47.5 GB | 39.3 GB | €4,255 used, ebay.co.uk | n/a |
| NVIDIA RTX A6000 48GB workstation GPU | 1 device | 47.5 GB | 39.3 GB | €4,349 used, ebay.ca | n/a |
| Apple Mac Studio M5 Ultra 96GB (2026, base Ultra) unified memory | 1 device | 72.0 GB | 63.8 GB | €4,784 new, apple.com | ≤ 203 tok/s |
| HP Z2 Mini G1a (Ryzen AI Max+ PRO 395, 128GB) unified memory | 1 device | 120 GB | 112 GB | €5,686 new, mindfactory.de | ≤ 43 tok/s |
| Lenovo ThinkStation PGX (GB10, 128GB) unified memory | 1 device | 120 GB | 112 GB | €5,829 new, karley.de | ≤ 46 tok/s |
| Apple Mac Studio M4 Max 128GB (2025) unified memory | 1 device | 96.0 GB | 87.8 GB | €5,979 used, ebay.co.uk | ≤ 92 tok/s |
| Apple Mac Studio M5 Max 128GB (2026) unified memory | 1 device | 96.0 GB | 87.8 GB | €6,189 new, macconsultshop.de | ≤ 104 tok/s |
| Dell Pro Max with GB10 (FCM1253, 128GB) unified memory | 1 device | 120 GB | 112 GB | €6,663 new, eurieka.ie, excl. VAT | ≤ 46 tok/s |
| Apple Mac Studio M3 Ultra 256GB (2025) unified memory | 1 device | 192 GB | 184 GB | €6,724 used, apple.com | ≤ 139 tok/s |
| Gigabyte AI TOP ATOM (GB10, 128GB) unified memory | 1 device | 120 GB | 112 GB | €6,795 new, grafikkarten.com | ≤ 46 tok/s |
| NVIDIA DGX Spark (GB10, 128GB) unified memory | 1 device | 120 GB | 112 GB | €6,999 new, grafikkarten.com | ≤ 46 tok/s |
| HP ZGX Nano G1n AI Station (GB10, 128GB) unified memory | 1 device | 120 GB | 112 GB | €7,127 new, office-partner.de | ≤ 46 tok/s |
| NVIDIA GeForce RTX 5090 32GB consumer GPU | 1 device | 31.5 GB | 23.3 GB | €7,262 new, alternate.de | n/a |
| MSI EdgeXpert MS-C931 (GB10, 128GB) unified memory | 1 device | 120 GB | 112 GB | €7,308 new, buyzero.de | ≤ 46 tok/s |
| NVIDIA RTX 6000 Ada 48GB workstation GPU | 1 device | 47.5 GB | 39.3 GB | €9,393 new, jacob.de | n/a |
| Apple Mac Studio M5 Ultra 256GB (2026) unified memory | 1 device | 192 GB | 184 GB | €9,830 new, abt.com | ≤ 203 tok/s |
| NVIDIA A100 80GB PCIe server GPU (passive cooling) | 1 device | 79.5 GB | 71.3 GB | €14,999 used, ebay.nl | n/a |
| NVIDIA DGX Station (GB300) unified memory | 1 device | 740 GB | 732 GB | €83,738 new, configurator.exxactcorp.com | ≤ 1202 tok/s |
| NVIDIA GeForce RTX 4080 Super 16GB consumer GPU | 1 device | 15.5 GB | 7.3 GB | no verified price | n/a |
| NVIDIA GeForce RTX 4070 Ti Super 16GB consumer GPU | 1 device | 15.5 GB | 7.3 GB | no verified price | n/a |
| AMD Radeon RX 7900 XTX 24GB consumer GPU | 1 device | 23.5 GB | 15.3 GB | no verified price | n/a |
| AMD Radeon Pro W7900 48GB workstation GPU | 1 device | 47.5 GB | 39.3 GB | no verified price | n/a |
| Apple Mac Studio M5 Ultra 512GB (2026) unified memory | 1 device | 384 GB | 376 GB | no verified price | ≤ 203 tok/s |
Speed ceiling is memory bandwidth divided by the bytes of (active) weights read per token, shown only where we have a sourced bandwidth figure. Real decode speed is lower, especially with long contexts or when splitting across cards. All memory numbers are estimates; see the methodology.
FAQ
- How much VRAM does Qwen3.5 9B need?
- About 8.2 GB at Q4_K_M with a 32K-token context (5.9 GB weights, 1.1 GB KV cache, 1.2 GB overhead). These are estimates.
- Can I run Qwen3.5 9B on a single GPU?
- Yes, at Q4_K_M and 32K context a NVIDIA Tesla V100 16GB (PCIe or SXM2) (16 GB) has enough memory, from about €275.
- Which apps can run Qwen3.5 9B locally?
- Verified so far: Ollama, LM Studio, llama.cpp, MLX (mlx-lm), KTransformers, vLLM, SGLang. Each entry links to its source on this page.